README.md

metadata

license: mit
language:
  - hy
  - en
tags:
  - art
  - poetry
  - Hovhannes_Tumanyan
  - Armenian
  - seq2seq
  - NLP-task
size_categories:
  - 10K<n<100K

📝 Seq2Seq Dataset: Hovhannes Tumanyan's Poems

This dataset contains sentence pairs extracted from the poetic works of Hovhannes Tumanyan, one of the most celebrated Armenian poets. It is formatted for training sequence-to-sequence (Seq2Seq) models for tasks such as:

• Text generation
• Dialogue modeling
• Style imitation

📂 Dataset Structure

The dataset is provided as a .csv file with two columns:

input_sentence	target_sentence
Line 1 of poem	Line 2 of poem
Line 2 of poem	Line 3 of poem
...	...

Each row represents a sentence pair where the input is a line from a Tumanyan poem and the target is the next consecutive line. This format is ideal for training models to learn the flow, rhythm, and style of Tumanyan’s poetic language.

🧠 Use Cases

• Train a chatbot that speaks in Tumanyan’s style.
• Generate poetry line-by-line.
• Fine-tune transformer models (e.g., GPT, T5, BART) on Armenian poetic structure.
• Build educational tools for Armenian language learning.

📜 Source

All poems are in the public domain and attributed to Hovhannes Tumanyan (1869–1923). Texts were cleaned and segmented line by line to maintain poetic continuity.

💡 Recommended Preprocessing

• Normalize punctuation (e.g., Armenian vs. Latin commas).
• Tokenize properly if training tokenizer-based models (e.g., HuggingFace tokenizers).
• Optionally, remove very short lines to avoid low-content samples.

📘 License

MIT

---